[IPv6]Re: [v6ops] New draft: "The IPv6 Loopback Address Prefix"

[Maciej Żenczykowski <[e-mail]>]

On Tue, Nov 25, 2025 at 4:50 PM Geoff Huston <[e-mail]> wrote:
>
> Hi Mark,
>
> I hear you, but I also observe that a prior effort to expand this designation
> in 2013 did not get beyoind a draft. This incarnation of the proposal is deliberately
> more modest (even then I see David Farmer saying (paraphrased) "whoa! Way
> too much!" So we have one comment saying "too big" and another saying "too
> small"!
>
>
> I'm inclined to say that the draft will keep this proposal  at a /96 for now but
> doubtless Warren and I will keep a close eye on any other comments
> that have a view about the appropriate size for a loopback prefix.
>
> But I do note:  "Those loopback interface addresses were all announced into
> the routing protocol" might bne interpreted as a bit contrary to RFC 4291
> which says quite  definitively "keep loopback em to yourself" So I am not
> exactly on board with a rationale that relates to a use scenario in routing
> protocol that "exports" these addresses out to neighbours, if I interpret your
> description correctly.
>
> regards,
>
>   Geoff

Whatever you do, don't make it smaller than the 24-bits we get with
127/8 for IPv4,
we don't need even more arguments for IPv6 being worse than IPv4.

So IMHO either ::/96 (so it doesn't hit ::ffff:a.b.c.d) or ::/104 (so
it matches the 24-bits of 127/8)
[except ::/128] seem reasonable.

In theory ::/81 would work too, as would ::/80 with an exception at
the end of the range for the final /96 of ::ffff:a.b.c.d/96
but I'm not sure such games are truly worth it - although more bits
are always nice if you need to randomly assign this stuff for whatever
reasons -- though at that point ULA might be better.

In practice I've seen a ULA /64 used for 'machine local' addressing
(covering multiple VMs on a single physical host, for example for
talking to the hypervisor's daemons, which are always at the same
machine-local-ula-ip:port on each host) which isn't really the same as
loopback anyway.

I do believe you MUST include ::1 in the range and make it a single
continuous a..b range (even if it isn't a full subnet), so I think
that's where it ends (ie. cannot go bigger than ::/80).